@InProceedings{SilvaGalvSant:2014:ReNeAp,
author = "Silva, Ricardo Dal'Agnol da and Galv{\~a}o, L{\^e}nio Soares and
Santos, Jo{\~a}o Roberto dos",
affiliation = "{Instituto Nacional de Pesquisas Espaciais (INPE)} and {Instituto
Nacional de Pesquisas Espaciais (INPE)} and {Instituto Nacional de
Pesquisas Espaciais (INPE)}",
title = "Redes neurais aplicadas ao estudo de florestas prim{\'a}ria e
secund{\'a}ria com dados espectral/textural ali/eo-1 / Neural
networks applied to the study of primary and secondary forests
with spectral/textural ali/eo-1 data",
booktitle = "Anais...",
year = "2014",
pages = "629--639",
organization = "Semin{\'a}rio de Atualiza{\c{c}}{\~a}o em Sensoriamento Remoto
e Sistemas de Informa{\c{c}}{\~o}es Geogr{\'a}ficas Aplicados
{\`a} Engenharia Florestal, 11. (SenGeF).",
publisher = "IEP",
address = "Curitiba",
keywords = "Florestas tropicais, sucess{\~o}es secund{\'a}rias, redes
neurais artificiais, ALI/EO-1, textura GLCM, Tropical forests,
secondary successions, artificial neural networks, ALI/EO-1, GLCM
texture.",
abstract = "As sucess{\~o}es secund{\'a}rias s{\~a}o tipologias importantes
para a manuten{\c{c}}{\~a}o da biodiversidade, regime
hidrol{\'o}gico e sequestro de carbono. A utiliza{\c{c}}{\~a}o
de m{\'e}tricas texturais GLCM pode colaborar na
discrimina{\c{c}}{\~a}o dessas classes por extrair a
variabilidade espacial do dossel florestal. Assim sendo,
tamb{\'e}m se faz necess{\'a}ria uma t{\'e}cnica como redes
neurais artificiais para sele{\c{c}}{\~a}o dos atributos mais
relevantes e integra{\c{c}}{\~a}o desses dados. O objetivo do
presente estudo foi de avaliar e comparar o uso de atributos
espectrais ALI/EO-1 e m{\'e}tricas texturais GLCM utilizando a
t{\'e}cnica de redes neurais artificiais Multi-Layer Perceptron
para mapeamento da cobertura da terra na Floresta Nacional do
Tapaj{\'o}s e arredores, com foco na discrimina{\c{c}}{\~a}o
das tipologias florestais prim{\'a}rias e sucess{\~o}es
secund{\'a}rias. Observou-se que os atributos texturais mais
relevantes foram a textura m{\'e}dia das bandas 3, 4, 6, 7 e 8, e
textura dissimilaridade da banda 8. Esses atributos, ao serem
integrados aos dados espectrais em um conjunto h{\'{\i}}brido,
proporcionaram uma melhor discrimina{\c{c}}{\~a}o entre as
classes de NPV e solo, culturas agr{\'{\i}}colas e SS1/SS2, SS1
e SS2, SS2 e SS3/FP. Dessa forma, as {\'a}reas de SS1, SS2, SS3 e
FP puderam ser discriminadas com 89, 63, 62 e 83% de
acur{\'a}cia. Constatou-se exatid{\~a}o global de 89% para a
utiliza{\c{c}}{\~a}o dos dados h{\'{\i}}bridos contra 79% para
dados somente espectrais. ABSTRACT Secondary successions are
important typologies for biodiversity maintenance, hydrological
regimen, and carbon sequestration. The use of GLCM textural
metrics can collaborate to discriminate these classes due to the
extraction of the spatial variability of the forest canopy. Hence,
it is also necessary a technique such as artificial neural
networks to select the most relevant attributes and to integrate
these data. The aim of this study was to evaluate and compare the
use of ALI/EO-1 spectral attributes and GLCM textural metrics
using the Multi-Layer Perceptron artificial neural networks
technique for land cover mapping in the Tapajos National Forest
and vicinity, focusing on the discrimination of primary forest and
secondary successions. It was observed that the most important
textural attributes were the mean texture of bands 3, 4, 6, 7 and
8, and the dissimilarity of band 8. These attributes, when
integrated into the spectral data to compose a hybrid dataset,
provided better discrimination between the classes of NPV and
soil, crops and SS1/SS2, SS1 and SS2, SS2 and SS3/PF. Thereby, the
SS1, SS2, SS3 and FP areas could be discriminated with 89, 63, 62
and 83% of classification accuracy. It was observed an overall
accuracy of 89% using the hybrid dataset against 79% using only
the spectral data.",
conference-location = "Curitiba",
conference-year = "14-16 out. 2014",
issn = "2178-8634",
label = "lattes: 6150479997891841 1 SilvaGalvSant:2014:ReNeAp",
language = "pt",
targetfile = "lenio redes.pdf",
urlaccessdate = "20 abr. 2024"
}